Build Strategies

Build is the part of Turbine Studio where an idea becomes an executable strategy. You describe the trading behavior you want, Studio asks for missing details, and the AI drafts a structured strategy that you can inspect before testing or running.

The goal is not to hide the strategy behind generated code. The goal is to make the strategy explicit enough that a human, an AI agent, and the runner can all reason about the same plan.

What you can build

Studio is designed for prediction market strategies such as:

  • spread capture and market making,
  • mean reversion after a price dislocation,
  • momentum or drift around scheduled events,
  • panic fades after fast moves,
  • data-driven rules using external signals,
  • position management and exit rules around market close,
  • strategy variants for Kalshi and Polymarket-style event markets where supported.

The first draft should be narrow. "Trade all weather markets profitably" is too broad. "Fade Chicago high-temperature YES contracts when NWS data implies a lower probability and the spread is under 5 cents" gives Studio enough structure to produce a testable draft.

The build loop

  1. Describe the thesis

    Name the market, the signal, the expected edge, and the failure condition.

  2. Choose the market scope

    You can ask for one market, a market family, or a selector such as a category or venue-specific identifier. Narrow selectors make backtests and deployments easier to reason about.

  3. Define entry logic

    Entry logic says when the strategy should open a position or post liquidity. It can reference market prices, spread, time, volume, and supported edge data.

  4. Define exit logic

    Exit logic says when the strategy should close, reduce, cancel, or stop adding exposure.

  5. Set risk limits

    Risk limits are first-class. A strategy should define max position, price bounds, maximum spend, stop conditions, and timing restrictions before it is backtested.

  6. Inspect the strategy

    Studio produces a structured strategy draft. Treat this as the contract between your intent and execution.

  7. Backtest

    Backtesting turns the draft into evidence. Good backtests do not only report upside; they reveal whether the strategy depends on unrealistic fills, stale data, or a tiny sample.

  8. Deploy

    Once you approve the behavior, Studio can deploy the strategy to your independent runner.

Good strategy prompts

Strong prompts include the market, signal, action, risk cap, and stop conditions:

Build a Kalshi BTC strategy that buys YES when Coinbase BTC spot rises more than 1.2% over 15 minutes while the market mid has moved less than 4 cents. Avoid spreads above 6 cents, cap exposure at $300, and exit if the signal reverses or the market enters the final 30 minutes.
Create a weather fade strategy for supported temperature markets. Use NWS observations and forecasts as the edge data. Sell overextended YES prices when the contract is more than 10 cents above the forecast-implied probability for 20 minutes. Do not trade if data is stale or the spread is wider than 5 cents.
Market-make a supported Polymarket event market with a minimum 3 cent spread, small order size, max inventory of $200, and automatic cancellation when the book gets thin or the market moves more than 8 cents in 5 minutes.

Weak prompts usually skip constraints:

Make me a profitable bot.
Trade elections aggressively.
Find arbitrage everywhere and run it.

Studio may still ask follow-up questions, but the better starting point is a scoped thesis with explicit limits.

What Studio asks for

When a prompt is underspecified, Studio may ask for:

  • venue or market family,
  • trade direction,
  • signal threshold,
  • maximum position,
  • price floors or ceilings,
  • minimum liquidity,
  • allowed holding period,
  • exit conditions,
  • whether the strategy should post maker orders or take liquidity,
  • whether the strategy may use edge data,
  • whether it should backtest before deployment.

AI agents should answer these questions directly instead of inventing hidden assumptions. If a human has not supplied a risk limit, the agent should ask for one or choose a conservative default and clearly label it.

The output

Build produces a strategy draft. A draft normally includes:

  • strategy type,
  • market selector,
  • data references,
  • loop cadence,
  • entry rules,
  • exit rules,
  • risk constraints,
  • deployment preferences,
  • backtest settings.

Studio can revise the draft repeatedly. The cleanest workflow is to change one behavior at a time, rerun the backtest, and compare the new result against the previous draft.

AI-friendly rules

If you are an AI agent using Studio:

  • Preserve the user's thesis in plain language.
  • Prefer explicit strategy changes over vague "improve performance" requests.
  • Never remove risk limits to improve a backtest.
  • Treat missing data as a reason to pause, not a reason to assume.
  • Backtest before proposing live deployment.
  • Explain trade-offs in terms of fills, exposure, liquidity, and overfitting.

Build is where the strategy becomes legible. A legible strategy is easier to test, easier to deploy, and easier to stop.